ServiceNow completed its acquisition of Element AI on January 8, 2021. All references to Element AI in the materials that are part of this project should refer to ServiceNow.

Probing Learning Algorithms with Synthetic Datasets

Description

This repository contains the code for reproducing experiments in [1].

To use the Synbols tool for generating new datasets, please visit https://github.com/ElementAI/synbols

Pytorch Dataset Quick Setup

from pytorch_examples.datasets import Synbols
from torchvision import transforms as tt

dataset_path = "./"
dataset_name = "default_n=100000_2020-Oct-19.h5py"

synbols = Synbols(args.data_path,
                  dataset_name=args.dataset)
train_dataset = synbols.get_split('train', tt.ToTensor())
val_dataset = synbols.get_split('val', tt.ToTensor())

For a complete example run ./pytorch_examples/minimal_classification.py from the root folder of this project:

python -m pytorch_examples.minimal_classification

It should reach >70% accuracy.

Bibliography

[1] Lacoste, A., Rodríguez, P., Branchaud-Charron, F., Atighehchian, P., Caccia, M., Laradji, I., Drouin, A., Craddock, M., Charlin, L. and Vázquez, D., 2020. Synbols: Probing Learning Algorithms with Synthetic Datasets.